foolen/

CogIP-Bench

CogIP-Bench is a comprehensive benchmark designed to evaluate and align Multimodal Large Language Models (MLLMs) with human subjective cognitive perception. While current MLLMs excel at objective recognition ("what is in the image"), they often struggle with subjective properties ("how the image feels"). This gap is what the CogIP-Bench aims to measure. This dataset evaluates models across four key cognitive dimensions: Aesthetics… See the full description on the dataset page: https://huggingface.co/datasets/foolen/CogIP-Bench.